Invitation to a talk by Prof. Yannis G. Kevrekidis (Chemical and Biological Engineering & Program in Applied Computational Mathematics at Princeton University) as part of the Kolloquium Thermo- und Fluiddynamik (KTF)
Date: Thursday, October 15, 2015
Location: HG D 7.1, ETH Zurich
I will first briefly review how matrix-free, and, in particular, simulator-based matrix-free nu-merical methods, can functionally link fine-scale simulators (MD, kMC…) with coarse-grained, systems-level numerical modeling tasks (like coarse stability and bifurcation com-putations). This is the “equation-free” component of the talk for coarse-graining large-scale, complex dynamical system models on the fly.
I will then discuss what we are currently concerned about/working on in my group: the use of data-mining techniques (the “variable-free” component) as part of the overall computa-tional process. Processing the results of brief bursts of multiscale simulations with tools like Diffusion Maps can suggest good macro-scale observables (variables) in terms of which macroscopic equations can in principle be written/solved. Performing scientific com-putation in terms of these “data-mining-based” variables poses a number of interesting problems that I will outline and discuss.
Host: Dr. C. Frouzakis, Prof. K. Boulouchos